Every respondent in the Oxford Economics and SAP survey of 300 senior customer experience (CX) executives is running some form of AI in their CX function. Having AI is no longer what separates the organizations getting results from the ones that aren’t.  

The separating factor, according to the research, is whether AI can access the data and processes it needs. Most organizations can’t say yes to that. Fragmented systems, disconnected data sources, and integration workarounds limit what AI can reason with, leaving many stuck using it for basic tasks or cycling through pilots that never scale.   

The study identified a small group of respondents—a segment it refers to as “leaders”—who are getting dramatically different outcomes than other organizations by investing in both platform harmonization and employee readiness. 

Matthew Reynolds, Senior Editor for Thought Leadership – Technology at Oxford Economics, presented the findings in a recent SAP webinar.  

The barriers respondents are facing are consistent: data quality and compliance concerns, integration challenges, and a lack of internal expertise. AI tool readiness is the lowest-ranked barrier. 

Just 25% of respondents describe their CX technology as fully harmonized. These CX leaders—defined by that harmonization—are significantly more likely to have AI embedded directly into enterprise workflows, and the performance gap that creates is stark.  

They call AI moderately or very effective in sales forecasting at 94% versus 54% of non-leaders, and in supply chain at 100% versus 55%. They ranked revenue as AI’s most positive impact at twice the rate of their peers (44% versus 22%). 

From Chatbots to Agents  

The next frontier in CX is agentic AI: systems that act on information, orchestrating processes like lead-to-cash, quote-to-order, and issue-to-resolution that span multiple business functions simultaneously. That kind of orchestration requires the agent to have access to inventory constraints, pricing logic, fulfillment timelines, and customer history in one place. 

Most organizations aren’t there yet. Nearly half are still exploring their first agentic use cases. But those who are excelling are in a different position, with 28% having deployed at scale—3.4 times the rate of their peers. 

These leaders rank workforce readiness and platform readiness as nearly equal prerequisites for AI success. “Tech infrastructure and employee readiness are two sides of the AI coin,” said Reynolds. But 23% of organizations are committing only minimal resources to training employees on the data and processes that those agents are working with.  

“There’s a back and forth that needs to happen, and that can be a little painful,” he noted. “It can be a little deflating at points. But getting through that first stage helps you figure out a process of how you can work across teams.” 

Those leaders who went through that process early are now seeing the payoff—78% report moderate or significant value from agentic AI in customer retention, versus 44% of non-leaders. 

The Blueprint  

Reynolds outlined four priorities for organizations looking to close the gap.  

• First, make platform harmonization a board-level conversation. “Any AI tool needs to be able to connect important apps, processes, systems that move away from fragmentation towards that harmony,” he said. “From a CX standpoint, CRM is the central focus.”

• Second, move from pilots to scale deployment in areas where AI is already proving itself. Sales forecasting, customer interaction management, and personalized marketing all showed large performance gaps between leaders and the rest of the field.

  Reynolds’s team found that transformation issues arise throughout the process, from pitch and validation through deployment and modification. “Finding the resources to carry out an AI transformation and getting the workforce to embrace new ways of working play significant roles in getting these efforts off the ground,” he said. Organizations that wait for perfect conditions don’t avoid those challenges. They just haven’t encountered them yet.  

• Third, start with compliance. This is counterintuitive, but Reynolds made a strong case for it. He pointed to compliance as a safe entry point for agentic deployment, in areas where agents can “insulate the business while also informing core business operations” and where, if something goes wrong, “nothing catastrophic happens.” The data supports the logic: leaders report industry-leading AI maturity in compliance at rates far beyond their peers (53% versus 8%).  

• Fourth, measure broadly. “You need to have a long list of KPIs,” Reynolds said. “AI can take value in many forms.” Integrated infrastructure tends to surface value where organizations weren’t looking. The research found leaders reporting significant ROI in areas like auditing customer interactions (31% of leaders versus 19% of others) and automating routine tasks and administration (47% versus 18%), alongside the more expected gains in sales forecasting and customer service.

  “Keep your options open because you have no idea what kind of benefits AI is going to have once it’s in play,” he encouraged. “It could affect areas in completely different departments or different regions.”  

CX as a standalone technology stack is reaching its expiration date. The organizations in this research that are getting the most from AI aren’t treating CX as a front-office function bolted onto everything else; they’re operating in environments where CX and the systems that run the business—ERP, supply chain, finance—share the same foundation.   

The research covered 10 industries evenly, and while the specifics of integration vary across them, the pattern held: harmonized infrastructure and workforce investment separated the leaders from everyone else.  

Organizations that have already built that foundation face a specific question: whether to let AI work with what’s already connected. The performance gap in this research suggests that most haven’t done that yet. But the leaders who have—the ones who paired harmonization and workforce readiness with AI that orchestrates across functions—are the ones whose results everyone else is still trying to pilot their way toward.